You Exude What You Eat: How Carbon-, Nitrogen-, and Sulfur-Rich Organic Substrates Shape Microbial Community Composition and the Dissolved Organic Matter Pool

ABSTRACT

Phytoplankton is the major source of labile organic matter in the sunlit ocean, and they are therefore key players in most biogeochemical cycles. However, studies examining the heterotrophic bacterial cycling of specific phytoplankton-derived nitrogen (N)- and sulfur (S)-containing organic compounds are currently lacking at the molecular level. Therefore, the present study investigated how the addition of N-containing (glycine betaine [GBT]) and S-containing (dimethylsulfoniopropionate [DMSP]) organic compounds, as well as glucose, influenced the microbial production of new organic molecules and the microbial community composition. The chemical composition of microbial-produced dissolved organic matter (DOM) was analyzed by ultrahigh-resolution Fourier transform ion cyclotron resonance mass spectrometry (FT-ICR MS) demonstrating that CHO-, CHON-, and CHOS-containing molecules were enriched in the glucose, GBT, and DMSP experiments, respectively. High-throughput sequencing showed that Alteromonadales was the dominant group in the glucose, while Rhodobacterales was the most abundant group in both the GBT and DMSP experiments. Cooccurrence network analysis furthermore indicated more complex linkages between the microbial community and organic molecules in the GBT compared with the other two experiments. Our results shed light on how different microbial communities respond to distinct organic compounds and mediate the cycling of ecologically relevant compounds.

IMPORTANCE Nitrogen (N)- and sulfur (S)-containing compounds are normally considered part of the labile organic matter pool that fuels heterotrophic bacterial activity in the ocean. Both glycine betaine (GBT) and dimethylsulfoniopropionate (DMSP) are representative N- and S-containing organic compounds, respectively, that are important phytoplankton cellular compounds. The present study therefore examined how the microbial community and the organic matter they produce are influenced by the addition of carbohydrate-containing (glucose), N-containing (GBT), and S-containing (DMSP) organic compounds. The results demonstrate that when these carbon-, N-, and S-rich compounds are added separately, the organic molecules produced by the bacteria growing on them are enriched in the same elements. Similarly, the microbial community composition was also distinct when different compounds were added as the substrate. Overall, this study demonstrates how the microbial communities metabolize and transform different substrates thereby, expanding our understanding of the complexity of links between microbes and substrates in the ocean.

INTRODUCTION

Marine phytoplankton are the main source of labile organic matter in the upper ocean (1–3), with up to 50% of the carbon fixed by photosynthesis being released via exudation, viral lysis, and predation into surrounding seawater as dissolved (DOM) and particulate organic matter (POM) (4–7). Heterotrophic bacteria metabolize these released organic molecules and convert a portion into new DOM compounds (i.e., ~5 to 7% of the microbially produced DOM are recalcitrant compounds with longer turnover time) (8–10), and due to this, they play important roles in the elemental cycles of carbon (C), nitrogen (N), and sulfur (S) in the ocean (11–13).

Most phytoplankton-derived DOM, such as oligomers and monomeric compounds, are labile and rapidly consumed by bacteria (10, 14). Monosaccharides are the direct product of the photosynthetic reaction, with glucose being the most abundant neutral sugar in the ocean, where it can fuel over 30% of the microbial production (15). Glycine betaine (GBT) and dimethylsulfoniopropionate (DMSP) are representative of N- and S-containing organic compounds, which are synthesized by phytoplankton and found throughout the ocean (16–18). In general, GBT is an important component of the dissolved organic nitrogen (DON) pool (19), whereas DMSP constitutes around 3 to 10% of the phytoplankton cell biomass and is part of the dissolved organic sulfur (DOS) pool (20). It has been reported that microorganisms can transform 41 to 91% of GBT carbon into cell biomass and respired CO₂ (21), suggesting that the remaining GBT carbon is released into the surrounding water as by-products. Similarly, DMSP can provide between 1 and 15% of the heterotrophic bacteria carbon (22–24) and the majority of their S demand (25, 26). Combined glucose, GBT, and DMSP are integral parts of the ocean C, N, and S biogeochemical cycles (27, 28), and understanding their interactions with microbes could potentially enhance our knowledge of their biogeochemistry.

Previous studies mostly focused on the bacterial response to in situ phytoplankton bloom conditions (28, 29) or how they react to the additions of complex organic matter derived from phytoplankton (30–32). The measurement, turnover, or degradation pathways of GBT and DMSP had also been elaborated (21, 26, 33). However, the molecular signatures of microbial metabolites derived from DMSP and GBT, and their impacts on microbial community require further investigation. Therefore, in this study, we determined how natural heterotrophic microbial responded to the addition of glucose, GBT, and DSMP by combining high-throughput DNA sequencing (amplicon sequencing of 16S rRNA gene fragments) with ultrahigh-resolution mass spectrometry (Fourier transform ion cyclotron resonance mass spectrometry [FT-ICR MS]). To avoid interferences from in situ background DOM and potential nutrient limitation, we conducted the experiments using artificial seawater and inoculated with natural bacterial community inoculum from the shelf surface seawater. Overall, our aim was to determine how different phytoplankton-derived organic substrates: (i) affect the microbial production of organic matter at the molecular level; (ii) affect the microbial community composition and succession; and (iii) reveal how this influences the interactions between the microbial community and organic molecules.

RESULTS

Variations of microbial abundances and nutrient concentrations.

The microbial abundance peaked on day 3 increasing from an initial ~2.2 ± 0.9 × 10⁸ to ~1.1 ± 0.1 × 10¹⁰ cells/liter in the glucose and GBT experiments (Fig. 1A). The microbial growth was slower in the DMSP experiment with abundance increasing from initial levels of ~2.2 ± 0.2 × 10⁸ to peak levels of ~3.8 ± 0.7 × 10⁹ cells/liter on day 15 (Fig. 1A). Following those peaks (day 3 for glucose and GBT and day 15 for DMSP), the microbial abundance generally declined until the end of the experiments. In all the experiments, the increases in microbial abundance generally mirrored a corresponding decrease in DOC concentrations (Fig. 1B). Similar trends were also found for the ammonium concentrations in the DMSP experiments, whereas it gradually increased until the end of the incubations (range from 1.2 to 48.2 μmol/liter) in the GBT experiment (Fig. S1A). The concentrations of sulfate remained stable in both the GBT (range from 25.2 to 25.3 mmol/liter) and DMSP (range from 165.7 to 171.5 μmol/liter) experiments due to the high background levels (Fig. S1B). Moreover, the concentrations of nitrate and nitrite remained relatively stable in the GBT (range from 1.55 to 1.65 μmol/liter) and DMSP (range from 61.5 to 62.2 μmol/liter) experiments over the entire experimental period (Fig. S1C). In the experiments, it was estimated that more than 80% of the N in GBT substrate was converted into ammonium, while ~10% of the S in DMSP was converted into sulfate.

FIG 1.

(A, B) Changes in the microbial abundances (A) and dissolved organic carbon (DOC) concentrations (B) in the three (glucose, glycine betaine [GBT] and dimethylsulfoniopropionate [DMSP]) experiments. (C) Relative changes of fluorescent intensity in four fluorescence components (C1, protein-like; C2 to C4, humic-like) in all experiments. The x axis (time) is log-transformed.

Dynamics of fluorescent DOM components.

A four-component model was obtained by parallel factor analysis (PARAFAC) analysis for the fluorescent DOM pool (Fig. 1C). It should be noted that in this article, we compare the relative changes of fluorescence intensity between thawed samples. Fluorescent DOM (FDOM) component C1 (excitation/emission [Ex/Em], 260/303 nm) resembled tyrosine-like fluorescence and commonly has been shown to increase in intensity during the microbial degradation of autochthonous labile organic matter (34). Component C2 (Ex/Em, 320/383 nm) represented humic-like fluorescence, which is commonly reported as a by-product of microbial organic matter degradation (35). The intensities of both component C1 and C2 generally increased throughout all experiments. Components C3 (Ex/Em, 355/450 nm) and C4 (Ex/Em, 260 [395]/480 nm) were similar to the humic-like fluorescence, which previously has been considered to be of terrestrial origin (36, 37). The fluorescence intensities of C3 and C4 in both the GBT and DMSP experiments showed an initial increase and decreased thereafter until the end of the experiment. In contrast, C3 increased in the glucose experiment, while C4 remained at very low and stable levels throughout.

DOM molecular characteristics and chemodiversity.

Generally, molecular richness (the number of identified formulas) increased during the experiments (Fig. 2). There was an increase of the average intensity-weighted ratios of oxygen to carbon (O/C^w), while the average intensity-weighted ratios of hydrogen to carbon (H/C^w) declined. The identified formulas consisted of carbon (C), hydrogen (H) and oxygen (O) (CHO), and with additional nitrogen (N, CHON), sulfur (S, CHOS) and NS (CHONS). The average relative intensity of CHO group formulas was higher in the glucose (73.2 ± 4.1%) compared to the GBT (61.4 ± 16.2%) and DMSP (59.6 ± 20.3%) experiments. The average relative intensity of CHON group formulas was slightly higher in the GBT (14.4 ± 5.0%) compared to the glucose (13.2 ± 3.1%) and DMSP (11.9 ± 1.5%) experiments. The average relative intensity of CHOS group formulas was higher in the DMSP (25.5 ± 14.6%) compared to the glucose (13.2 ± 1.5%) and GBT (20.6 ± 11.2%) experiments. A molecular lability boundary (MLB) was applied by dividing the formula richness into more labile (MLB_L, H/C ≥ 1.5) or more recalcitrant (MLB_R, H/C < 1.5) contributions (38). The MLB_L peaked on day 7 in all experiment, suggesting that the microbes produced labile molecular species through the utilization of the three substrates. The MLB_L decreased, while the MLB_R increased until the end of the experiments, indicating that further microbial transformation resulted in more recalcitrant molecular species.

FIG 2.

Molecular characteristics of dissolved organic matter (DOM) identified by ultrahigh-resolution mass spectrometry in the three (glucose, glycine betaine [GBT], and dimethylsulfoniopropionate [DMSP]) experiments. Here, we show richness (the number of identified formulas), the average intensity-weighted ratios of hydrogen to carbon (H/C^w) and oxygen to carbon (O/C^w), relative intensity of different elementary formulas (CHO-, CHON-, CHOS-, and CHONS-containing formulas), and the percentage of more labile (molecular lability boundary [MLB], MLB_L, H/C ≥ 1.5) and more recalcitrant formulas (MLB_R, H/C < 1.5). The x axis (time) is log-transformed.

Subsequently, based on the presence and absence of molecular formulas, we defined the intermediated molecular species (newly produced and then degraded), remaining molecular species (newly produced and remained until the end of the experiment), and unique molecular species (newly produced and existed only in one substrate experiment) in the three experiments (Fig. S2). The MLB_L was higher in the intermediated molecular species in all experiments, which was consistent with a more labile nature of the compounds that could be degraded during the experiments. On the contrary, the remaining molecular species consisted of higher MLB_R in accordance with a recalcitrant nature that persisted until the end of the experiments. In addition, for the unique molecular species, a higher proportion of CHO group formulas were present in the glucose (35.2%) compared to GBT (22.9%) and DMSP (10.4%) experiments. The GBT (51.8%) experiment resulted in a higher proportion of CHON group formulas in the unique molecular species compared to the glucose (47.3%) and DMSP experiments (9.1%). A higher proportion of CHOS group formulas of the unique molecular species were found in the DMSP (73.5%) compared to the glucose (15.0%) and GBT (14.9%) experiments. Further, unique CHON, CHONS, and CHOS species were characterized as GBT-derived (newly produced in the GBT experiment) and inorganic N-derived (newly produced in the glucose and DMSP experiments) N-containing formulas and DMSP-derived (newly produced in the DMSP experiment) and inorganic S-derived (newly produced in the glucose and GBT experiments) S-containing formulas (Fig. S3). The MLB_L values were higher in both the GBT-derived (56.9%) and DMSP-derived (74.2%) formulas compared to the inorganic derived formulas (inorganic N, 24.3%; inorganic S, 49.7%). This result suggests that organic N/S (i.e., GBT/DMSP)-derived formulas contributed more to the labile DOM pool, while the products of inorganic N/S displayed relatively more recalcitrant properties.

Microbial community composition and clustering analysis.

To gain insight into changes in the microbial community composition, a total of 971 unique taxa with high taxonomic resolution as amplicon sequence variants (ASVs) were characterized from the 16S rRNA gene sequencing. Over the entire period and for all experiments, the major bacterial groups were Rhodobacterales (37.4%), Alteromonadales (18.9%), and Flavobacteriales (14.5%) (Fig. 3A). In the glucose experiment, Alteromonadales contributed substantially to the overall community on the third day, after which the relative abundance gradually decreased. This decline was followed by an increase of Rhodobacterales and Flavobacteriales until the end of the experiment. In the GBT experiment, Rhodobacterales and Oceanospirillales responded quickly, reaching a maximum contribution on the third day, which was succeeded by an increasing relative contribution of Flavobacteriales. In the DMSP experiment, relative abundance of Alteromonadales and Oceanospirillales peaked on the third day, while Rhodobacterales made up a large proportion (ranging from 71.7 to 98.2%) of the community from days 3 to 30. Generally, the microbial Shannon and Pielous indices (diversity and evenness) decreased until day 3/7, suggesting that opportunistic taxa initially dominated the microbial communities. Following this, the Shannon and Pielous indices increased until the end of the experiments, indicating the growth of a divers microbial community (Fig. S4).

FIG 3.

Microbial community composition (A) and cluster analysis (B) based on order and amplicon sequence variant (ASV) levels, respectively, over the experimental period (days 0, 3, 7, 15/16, 30, and 110/120) with the three substrates (glucose, glycine betaine [GBT], and dimethylsulfoniopropionate [DMSP]).

A cluster dendrogram was constructed to further analyze the similarities and differences of the microbial community composition across experiments (Fig. 3B). The lower height value indicates higher similarity. The two original microbial communities added to the substrates were grouped into the same cluster, suggesting that the different natural communities added to the glucose and the GBT and DMSP experiments were more similar compared to the subsequent microbial communities growing in the three experiments. Over the course of the experiments, the microbial communities started to cluster with the initial inoculum of microbial assembly first in the GBT (day 7), followed by the glucose (day 30), and lastly on day 120 in the DMSP incubations.

Linking major microbial groups with DOM molecular characteristics.

A redundancy analysis (RDA) was conducted to assess the effect of different molecular characteristics in shaping the major microbial orders (Fig. 4). The first axis contributed to 37.5% of the explained weights, which was significantly (analysis of variance [ANOVA], P < 0.01) correlated with the substrates (Fig. S5). The second axis contributed to 10.3% of the explained weights. Along the negative first axis, Alteromonadales was associated with the microbial community in the glucose experiment. In contrast, along the positive first axis, Rhodobacterales was linked to the microbial communities in the GBT and DMSP experiments. Flavobacteriales was found to be related to microbial communities in all three experiments along the positive second axis. In addition, Alteromonadales was positively correlated with MLB_R. Rhodobacterales showed positive correlation with the protein-like C1 component, as well as H/C^w and CHOS group formulas. Flavobacteriales was found to be positively related to the humic-like C2 component and CHON formulas.

FIG 4.

Redundancy analysis (RDA) plots derived from major microbial orders and molecular characteristics (the average intensity-weighted ratios of hydrogen to carbon [H/C^w], relative intensity of different elementary formulas [CHON- and CHOS-containing formulas], fluorescence components [C1, protein-like; C2, humic-like], and the percentage of more recalcitrant formulas [MLB_R, H/C < 1.5]) in the three substrates (glucose, glycine betaine [GBT], and dimethylsulfoniopropionate [DMSP]).

Association between the microbial community and DOM formulas.

Network analysis was used to describe cooccurrence between individual microbial taxa and organic molecules and to analyze the potential microbial-organic molecular interactions among the three different experiments (Fig. 5A). The corresponding ASVs included up to 69.8, 36.8, and 29.5% (relative abundance) of the total community in the networks, and the networks consisted of up to 3.0, 9.1, and 14.7% (relative intensity) of the formulas in the glucose, GBT, and DMSP experiments, respectively (Table S1). The networks included more ASVs and formulas in the GBT (63 and 198) compared to the glucose (42 and 92) and DMSP (38 and 172) experiments. Further, the degree number, indicating the connections between ASVs and formulas, increased from the glucose (131) to the DMSP (312) and GBT (809) experiments. In addition, there was also a higher proportion of labile formulas in the GBT (44.4%) compared to the glucose (34.8%) and DMSP (29.1%) networks. These results suggest that more numerous and labile formulas correlated with additional ASVs so that more potentially microbial-organic molecular associations occurred in the GBT network (Fig. 5A; Table S1). Furthermore, some distinct ASVs in the networks showed divergent microbial-organic molecule associations. For example, ASV376 and ASV386 belonging to Alteromonadales connected with more CHO group formulas of a less labile nature (Fig. 5B). Conversely, ASV514, ASV346, and ASV585 of Alphaproteobacteria, Rhodobacterales, and Flavobacteriales, respectively, linked to more CHON and CHOS group formulas with a more labile nature.

FIG 5.

Network analysis based on the Spearman correlation between the microbial amplicon sequence variants (ASVs) and formulas identified by ultrahigh-resolution mass spectrometry in the three substrates (glucose, glycine betaine [GBT], and dimethylsulfoniopropionate [DMSP]). (A) The Spearmen’s rank correlation coefficient is strong when r² = 1, P < 0.01, and Q value < 0.05. ASVs and formulas appearing in at least half of the samples were included in the network analysis. Subnetworks a to c represent microbes and formulas presented only in the glucose, GBT, and DMSP groups, respectively. Subnetworks d to f represent microbes and formulas shared by two groups, respectively. Subnetwork g represents microbes and formulas presented in all three groups. The color bars represent the different microbial taxonomy and elemental formulas. The diamond and arrow shapes represent more labile (molecular lability boundary [MLB], MLB_L, H/C ≥ 1.5) and more recalcitrant formulas (MLB_R, H/C < 1.5). The color lines represent the positive (red) and negative (blue) relationship. (B) Top five ASVs with high degree (number of the connections between ASVs and formulas) and associated formulas from the networks are shown.

DISCUSSION

The present study investigated the microbial transformation of distinct organic compounds (glucose, GBT, and DMSP), which allowed us to understand how different organic molecules shape the microbial community composition and vice versa (Fig. 6). Our results demonstrate that the addition of a N-containing compound (GBT) is more likely to maintain the microbial community composition and result in more complex associations between microbial composition and the organic molecules compared with the carbohydrate (glucose) or S-containing (DMSP) compounds.

FIG 6.

Diagram of how phytoplankton-derived photosynthate (glucose, glycine betaine [GBT], and dimethylsulfoniopropionate [DSMP]) fuel the microbial food web process and subsequently contribute to dissolved organic carbon (DOC, here mainly represents CHO-containing organic molecules), nitrogen (DON, including CHON- and CHONS-containing organic molecules), and sulfur (DOS, including CHOS- and CHONS-containing organic molecules) pools. Glucose, GBT, and DMSP are three important phytoplankton-derived labile organic compounds and represent carbohydrate-, N-containing, and S-containing substrates, respectively. During degradation of these labile organic matter, heterotrophic bacteria, especially Alteromonadales (dominant group in the glucose) and Rhodobacterales (dominant group in both GBT and DMSP), drive the C (carbon), N (nitrogen), and S (sulfur) elemental cycles.

The glucose experiment was dominated by Alteromonadales, while Rhodobacterales was the most abundant group in the GBT and DMSP experiments. Distinct dominating microbial groups responding to different substrates could be due to different microbial life strategies or metabolism capacities (39, 40). Members of Alteromonadales, Rhodobacterales, and Flavobacteriales are known as opportunists having diverse genomic and metabolic potentials, commonly making them the most dominant group especially in conditions with sufficient supply of labile organic material such as that found during phytoplankton bloom conditions (28, 41, 42). In our glucose experiment, there clearly were sufficient organic and inorganic nutrients, which resulted in the flourishing of opportunistic Alteromonadales (43). The success of Alteromonadales in utilizing glucose could be due to the phosphoenolpyruvate carboxylase strategy in Gammaproteobacteria instead of the pyruvate carboxylase in Alphaproteobacteria (44), giving them a competitive advantage. In contrast, the degradation of the more complex GBT and DMSP substrates requires specific transporters and enzymes (e.g., betaine-homocysteine methyltransferase and DMSP-dependent demethylase A protein), resulting in the dominance of Alphaproteobacteria (mostly Rhodobacteraceae) with a smaller contribution of the Gammaproteobacterial members (45–47). Previous studies have also shown that Alphaproteobacteria can account for up to 30% of total bacteria assimilating DMSP in natural environments (48). Usually Gammaproteobacteria, especially copiotrophic Alteromonas, benefits from the “bottle effect” (49–51) as we observed in the glucose experiment. Conversely, our experiments with GBT and DMSP supported the growth of Rhodobacterales. Additionally, a recent study has demonstrated that different microorganisms (Spongiibacteracea, Euryarcheaota, Thaumarcheaota, Alteromonas, and SAR11) dominate in experiments enriched in deoxycholate and benzoic acid (52). This suggested the different substrates determine distinct outcomes of the bottle effect. These precedent responders can release a wide range of metabolic by-products that influence the successive microbial community composition (53, 54). The succession of the microbial community differed in the three experiments, which is likely due to the different metabolism and release of distinct organic compounds that favor different groups. Similar results have recently been obtained in which the DOM released from two bacteria were shown to stimulate distinct microbial exoenzymatic activities and microbial communities (55). Moreover, microbial groups resembling the microbial communities of the initial inoculum first occurred in the GBT experiment, followed by glucose and DMSP experiments. These recurring patterns in the microbial community previously have been observed in coastal blooms (56) and laboratory incubations (57). Overall, our results demonstrate that different substrates determine distinct microbial community structure and the importance of N-containing organic substrates in maintaining the microbial community compositions.

The initial microbial community responded differently toward the three substrates, also leading to differences in the chemical composition of the microbial-produced organic matter. The production of protein-like FDOM component C1 was higher in the DMSP and GBT compared to the glucose experiment. This is likely due to the microbial degradation and transformation of DMSP-derived sulfur into proteins (33) and glycine, the by-product of GBT degradation, that could be potentially biosynthesized into tyrosine (58). The microbially produced humic-like fluorescent components are commonly transformed from labile DOM and shown in our experiments, in accordance with previous studies, longer wavelength fluorescence (59–61). In addition, we found that the humic-like fluorescence components C3 and C4 were produced during the initial phase of the GBT and DSMP experiments and then degraded, while C3 increased and C4 showed minor changes in the glucose experiment. These different patterns could be related to differences in the chemical composition and bioavailability of the produced compounds (62) and/or differences in the microbial population successions in the three experiments, because it has also been shown recently that there are strong linkages between DOM optical properties and different clades of bacteria (63). In line with this, the chemical characterization also showed that the number of ASVs and formulas all generally increased from day seven until end of the experiment. This also suggests that the production of new organic molecules enabled the growth of new microbial groups and vice versa as previously demonstrated in field studies (64–66).

Network analysis visualized the cooccurrence patterns between the microbial community and DOM molecules and provided insights of the potential interactions (66, 67). We found increasing numbers of correlated ASVs and organic molecules and more complex associations of microbial-DOM in the GBT compared to glucose and DMSP experiments. These results show the importance of N-containing substrate in strengthening the linkages between microbial community and molecular formulas. Recently, other studies have shown the microbial preference of CHON over CHOS and CHO molecules in coastal waters (68), which is also consistent with the preference for N-containing compared with CHO-containing organic molecules found in natural marine waters (69). Additionally, previous experiments have shown strong linkages between microbial community and N-containing molecules of Synechococcus lysates (67, 70). Therefore, N-rich substrates generally appear to be more important in fueling bacterial activity and shaping the community composition, as well as enhancing the organic molecular transformation and microbial-organic matter interactions. Here, it should be noted that only one N-containing photosynthate compound was tested in our experiment, and further studies are needed to verify these results.

In addition, microbial transformation of glucose, GBT and DMSP resulted in a higher proportion of produced CHO, CHON, and CHOS group formulas, respectively. The Alteromonadales thrived, especially in the glucose experiment, showing positive association with MLB_R. As previously described, members of Alteromonadales can produce condensed hydrocarbon-like formulas from a broad range of carbohydrates and they enable the consumption of labile DOC (50, 71). In contrast, Rhodobacterales was positively associated with S-containing formulas and relative labile protein-like C1 component and negatively associated with MLB_R. This fits well with previous findings demonstrating links between members of the Rhodobacteraceae family and sulfur metabolites (47, 72), but it also suggests that Rhodobacterales might be less competitive when the substrates are relatively more recalcitrant. Glucose assimilation results in biomass and energy production but also in the synthesis of other metabolic intermediates (e.g., amino acids and fatty acids) needed for bacterial growth (73, 74). The substrates GBT and DMSP can also serve as carbon sources, but they importantly also provide N and S, respectively, with each influences the microbial communities differently (25, 75). The unique organic N/S (GBT/DMSP)-derived formulas had a more labile nature compared to those produced from inorganic N/S, which is likely due to differences in the microbial biosynthesis and catabolism processes involved in producing the metabolites (76, 77). Our results indicate that three important phytoplankton photosynthates (C-, N-, and S-rich) play different roles in regulating ocean biogeochemistry and further shed light on the connectivity and complexity of marine organic C, N, and S pools and the bacteria community.

In conclusion, this study demonstrates that (i) chemical composition of different substrates determine distinct microbial growth and affect microbial community structure; (ii) the molecular composition of microbial-produced organic matter is constrained by the substrate; (iii) our network analysis revealed that N-rich substrate maintained more complex microbial-organic matter molecular associations compared to C- or S-rich compounds. These results shed light on the links among substrates-microorganism-metabolites and improve our understanding of the complexity of these connections.

MATERIALS AND METHODS

Experimental setup.

We prepared artificial seawater (ASW) to investigate the microbial degradation and transformation of GBT and DMSP, respectively. The ASW used in this study was modified from Kester et al. (78), comprising NaCl (24 g liter⁻¹), KCl (0.67 g liter⁻¹), KBr (0.1 g liter⁻¹), NaF (0.04 g liter⁻¹), H₃BO₃ (0.03 g liter⁻¹), CaCl₂ (0.05 g liter⁻¹), NaHCO₃ (0.2 g liter⁻¹), and KH₂PO₄ (2 mg liter⁻¹) dissolved in Milli-Q water (Fig. S6). We added MgSO₄·7H₂O (5 g liter⁻¹) in the ASW for the GBT experiment that had no inorganic nitrogen, while MgCl₂·6H₂O (4.12 g liter⁻¹), NaNO₃ (6 mg liter⁻¹), and NH₄Cl (4 mg liter⁻¹) were added in the ASW for the DMSP experiment without the addition of inorganic sulfur. Four salts (NaCl, KCl, KBr, and NaF) were precombusted (450°C, 6 h) before use. Two different organic substrates, GBT (98%, Sigma-Aldrich) and DMSP (98%, TCI), were prepared at final concentrations of ~300 μmol C/L in triplicate.

Shelf surface seawater was collected from station 41 (21.450N, 113.927E) in the northern South China Sea (Fig. S6). The seawater was filtered through a 3-μm polycarbonate filter. The seawater inoculum was diluted in a 1:50 ratio in part to reduce the influence of the in situ background DOM. Six 10-L polycarbonate bottles (acid washed and rinsed with Milli-Q water) were incubated for 120 days in the dark at room temperature (26°C) with subsamples collected from these bottles at six time points (days 0, 3, 7, 15, 30, and 120). The results are presented as averages (± standard deviation) of three replicates.

In this study, we compare the results obtained from the GBT and DMSP experiments with results from a similar experiment in which glucose was added to ASW seawater and inoculated with a natural microbial assembly collected near Xiamen Island (79). Therefore, the GTB and DMSP were distinctly compared with the glucose experiment, as the inoculated microbial assembly was different.

Sample collection.

In the experiments, unfiltered water from the incubation bottles was used to follow changes in microbial abundance and microbial community composition, while samples for the analysis of DOC concentrations, fluorescent DOM (FDOM), and DOM extraction were filtered through precombusted (450°C, 6 h) GF/F filters (nominal pore size, 0.7 μm; Whatman). Samples for microbial abundance (2 mL) were collected and fixed with glutaraldehyde (1% vol/vol) and kept at −20°C until analysis. Additionally, 300 mL of unfiltered water were filtered onto a 0.22-μm polycarbonate filter (Millipore, Billerica, MA, USA) for microbial community analysis. Subsequently, duplicate 40-mL filtrates were collected in 50-mL sterile tubes and stored at −20°C for inorganic nutrient analysis (ammonium and sulfate). Filtered samples (20 mL) for DOC and FDOM determination were stored in two different 40 mL glass vials at −20°C until analysis. For the DOM extraction, 360 mL filtrate was acidified to a pH of 2 in a 1-liter glass bottle and thereafter extracted (see details below). All glassware used in this study was acid washed, rinsed with Milli-Q water, and precombusted (450°C, 6 h) before use.

Sample analysis.

Ammonium concentrations were measured using the orthophthaldialdehyde fluorometric method (80). Sulfate concentrations was determined in a Dionex ICS-5000 ion chromatography system equipped with a chemical conductivity suppressor. The DOC concentration was analyzed using a Shimadzu TOC-VCPH analyzer. The samples were defrosted and acidified (pH 2) with phosphoric acid, after which the DOC concentrations were determined from the slope of a standard curve made from potassium hydrogen phthalate. The consistency of the results was determined by the stable measurement of Milli-Q water every sixth sample.

Spectral fluorescence was determined using a Horiba Aqualog system as previously described (66). Briefly, water samples were defrosted and allowed to warm to room temperature before scanning from 240 to 600 nm (every 5 nm) using a 1-cm-path length quartz cuvette. Emission spectra were recorded ranging from 248 to 829 nm (every 2.33 nm) at a 2-s integration time. All measured samples were calibrated to Raman units (RU) by the signal of Milli-Q water measured on the same day (excitation at 350 nm, emission from 371 to 428 nm) (81). Parallel factor analysis (PARAFAC) was performed to decompose combined excitation (Ex)-emission (Em) matrices (EEMs) into individual fluorescent components using the software MATLAB (82).

DOM extraction procedure was performed as previously described (83). Briefly, the filtered and acidified (pH 2) sample was passed by gravity through a cartridge (200 mg, Agilent Bond Elut PPL, USA) following the manufacturer’s guidelines. Eluted DOM was analyzed using a 9.4 T Bruker Apex Ultra with an Apollo II electrospray ion source operated in negative mode (84). The molecular identification was based on the elemental combinations of ¹²C₁₋₆₀, ¹H₁₋₁₂₀, ¹⁴N₀₋₃, ¹⁶O₀₋₃₀, and ³²S₀₋₁. Further, the identified formulas were assigned into different molecular groups (CHO-, CHON-, CHOS-, and CHONS-containing formulas), and thereafter formulas that appeared in at least two replicates were selected for further analysis. We calculated individual molecular relative abundance by dividing each molecular intensity by the sum of molecular intensity in the same sample for each time point, and average intensity-weighted ratios of hydrogen to carbon (H/C^w) and oxygen to carbon (O/C^w) were then calculated for each sample (85). The results are presented as averages of three replicates.

For the analysis of microbial abundance, thawed samples were stained with the nucleic acid-specific dye SYBR green I (Invitrogen, Carlsbad, CA, USA) (86) for 15 min in the dark and thereafter analyzed using a flow cytometry (BD Accuri C6). The phenol-chloroformisoamyl alcohol method was applied to extract microbial DNA as previously described (87). For the taxonomic identification of microbial compositions, the V4-V5 region (forward primers 515F 5′-GTGCCAGCMGCCGCGGTAA-3′ and reverse primers 907R 5′-CCGTCAATTCMTTTRAGTTT-3′) of the small ribosomal subunit (16S rRNA) gene were amplified using the PCR procedure (88). Quantified amplicons were sequenced using the Illumina MiSeq platform (Illumina, San Diego, CA, USA; for details, see supplemental material).

Statistical analysis.

The microbial community composition and molecular formulas were presented by the average relative abundance/intensity of three replicates, respectively, for further analysis. α-Diversity was determined by the Shannon and Pielous indices (89) using the package “picante.” Cluster analysis was visualized in a dendrogram to compare the similarity of the microbial community composition (ASV level) based on Euclidean distance. Redundancy analysis (RDA) was applied to access the microbial population successions (order level, normalized with the “Hellinger” method) with molecular characteristics (log transformed). The RDA were performed using the package “vegan” with the corrections of variance inflation factors (VIF) < 20 as a threshold. The RDA axis were tested by ANOVA. The associations between the individual ASVs and molecular formulas (both ASVs and formulas appeared in at least half the samples were selected) are presented in the cooccurrence network based on a strong Spearmen’s rank correlation coefficient of r² = 1, P < 0.01. Corresponding R codes were derived from GitHub (https://github.com/ryanjw/co-occurrence) (90). The multiple testing of the correlations was corrected (Q value < 0.05) using the package “fdrtool.” All of the statistical analyses were performed in R 3.6.1 (www.R-project.org), and network visualizations were conducted in Cytoscape version 3.7.2.

Data availability.

The sequence data were deposited in the National Center for Biotechnology Information (NCBI) Sequence Read Archive with BioProject PRJNA764558. FT-ICR MS data were deposited on figshare (https://doi.org/10.6084/m9.figshare.16640698.v3).